Undefined function 'conv2' for input arguments of type 'double' and attributes 'full 3d real'. - Matlab

Question

I'm trying to filter an image in the space domain so i'm using the conv2 function.

here's my code.

cd /home/samuelpedro/Desktop/APIProject/

close all
clear all
clc

img = imread('coimbra_aerea.jpg');
%figure, imshow(img);

size_img = size(img);

gauss = fspecial('gaussian', [size_img(1) size_img(2)], 50);

%figure, surf(gauss), shading interp

img_double = im2double(img);

filter_g = conv2(gauss,img_double);

I got the error:

Undefined function 'conv2' for input arguments of type 'double' and attributes 'full 3d
real'.

Error in test (line 18)
filter_g = conv2(gauss,img_double);

now i'm wondering, can't I use a 3 channel image, meaning color image.

supyo · Accepted Answer · 2012-12-15 02:12:09Z

11

Color images are 3 dimensional arrays (x,y,color). conv2 is only defined for 2-dimensions, so it won't work directly on a 3-dimensional array.

Three options:

Use an n-dimensional convolution, convn()
Convert to a grayscale image using rgb2gray(), and filter in 2D:

filter_g = conv2(gauss,rgb2gray(img_double));

Filter each color (RGB) separately in 2D:

filter_g = zeros(size(im_double));
for i = 1:3
  filter_g(:,:,i) = conv2(gauss, im_double(:,:,i);
end

edited Dec 15, 2012 at 2:12

answered Dec 15, 2012 at 1:38

supyo

3,0372 gold badges24 silver badges35 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

SamuelNLP Over a year ago

Filtering each color separately is a great idea. i didn't have thought of that. :)

Alex Taylor Over a year ago

If you want to do this faster, passing a multi-plane image to imfilter or since this question is doing gaussian filtering, to imgaussfilt, will be faster. Both of these functions know how to operate on multi-plane data without the use of a for loop. See my answer below for more info.

Jonas · Accepted Answer · 2012-12-15 01:13:46Z

1

For n-D input, you need to use convn.

answered Dec 15, 2012 at 1:13

Jonas

75.1k10 gold badges139 silver badges180 bronze badges

2 Comments

SamuelNLP Over a year ago

well, i was afraid that would be the only chance. but convn turns matlab really slow.

Jonas Over a year ago

@SamuelNLP: You can use convn, but with two 1-D Gaussian filters (one along X, the other along Y). That will speed up your code a lot.

Alex Taylor · Accepted Answer · 2015-10-06 02:18:31Z

1

If you have R2015a or newer, the IPT function imgaussfilt handles multi-plane 2-d convolution problems like this, just pass in your RGB image.

http://www.mathworks.com/help/images/ref/imgaussfilt.html

If you don't, imfilter also performs multiplane 2-d convolution.

Both will be faster for a Gaussian filter, they both know how to do the separable trick for you.

answered Oct 6, 2015 at 2:18

Alex Taylor

1,4121 gold badge9 silver badges15 bronze badges

1 Comment

Alex Taylor Over a year ago

Also remove the im2double when using imgaussfilt/imfilter, both apply SIMD optimizations for integer inputs, so they'll be faster without the cast.

Collectives™ on Stack Overflow

Undefined function 'conv2' for input arguments of type 'double' and attributes 'full 3d real'. - Matlab

3 Answers 3

2 Comments

2 Comments

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

2 Comments

2 Comments

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related